Ground-truthing requirements (Physical and biological samples)

The ground-truth survey should fulfil two roles. Firstly it should provide the biological and physical information required to determine or identify a habitat class. In general, so called ‘direct’ mapping exercises use ground-truth samples to assign a single habitat class to the ground-type and/or mapped polygon from which the samples were taken. It is not intended that the ground-truth sampling should greatly influence the segmentation of the survey area; its function is solely to determine the habitat class for areas that have already been mapped by direct interpretation of remote sampling images (aerial photographs or acoustic mosaics). In this way, direct mapping is akin to ‘painting by numbers’ as it imposes an attribute on a pre-defined area. However, in ‘modelled’ mapping, the initial segmentation of the survey area merely directs the placement of the ground-truth sampling effort. Once the samples have been collected the initial segmentation is discarded and an integrated analysis is made of the biological, physical and ‘coverage’ data (remote sensing and/or spatially modelled data) to produce a classified map. In this way, the ground-truth data influence both the classification and delineation of the habitats. 'Direct mapping' & 'Statistical modelling' give fuller accounts of these different approaches to mapping, but both require fundamental information on the physical nature of the substrate and the biological communities it supports, as detailed in 'Physical information' and 'Biological information'.

Secondly, ground-truth surveys provide the opportunity to verify the validity, nature and location of any of the putative borders between ground-types, segments or habitats that have been determined prior to the survey, and this requires some form of observational sampling. For both vector and raster type maps, checks should be made that the borders relate to real habitat changes on the ground rather than artefacts of the remotely sensed data or the mode of its interpretation. Patchy cloud cover can cause differential shading on aerial photographs and changes in slope may cause similar shading in sidescan images, but neither reflects a real change in habitat type on the ground. The very act of mosaicing images can also introduce false ‘shade’ borders (see the orthophotographs in the previous section ‘Optimise the ground-truthing’). Where an image shows a gradation from one ground type to another, ground observation can help to determine the appropriate placement and type of border (discrete or transitional) between different habitats.

Sampling effort

The amount of ground truthing effort required is influenced by several factors including, the heterogeneity of the study area, the required degree of confidence in the map and the level of detail required in the habitat classification.
Clearly a heterogeneous area with more ground types will need more sampling than a homogeneous area with few ground types. Also, if a ground type itself is found to be heterogeneous (i.e. a mixed substrate) then this will need more sampling effort to determine the range of habitats it contains (see also ‘Optimal Allocation Analysis’).
More samples tend to increase confidence in a map, as they reduce uncertainty in classification accuracy. Assigning a habitat class on the basis of a single sample will be less certain than assigning one based on, say, five samples that all show the same characteristics. Consequently, greater sampling density is needed in studies that require high levels of confidence in the mapped output. Four basic sampling regimes have been discussed earlier in the section ‘Ground-truth survey strategy’, and 'How good is my Map?' explains more about assessing confidence in maps.
It is the nature of hierarchical classification systems to use different types of information at different levels of the classification, so the more precise the classification, the more information is required (see the earlier summary table of EUNIS levels 3, 4 & 5). Some levels in the hierarchy require only physical information while others require both physical and biological information, which may lead to additional sampling requirement. However, if one level merely requires more detail than its predecessor (e.g. identifying organisms to species rather than Family level), this can be satisfied by more detailed analysis of existing samples.
Biological assemblages comprise both infuana and epifauna (and flora) and both must be sampled to provide a complete description of the habitat. A single sampling event or device is rarely adequate for sampling both faunal fractions, so it is normal to sample each fraction separately using gear and protocols specifically designed for the job (see How do I collect my data?).

Directed ground-truthing.

Ground-truth techniques can differ greatly between inter-tidal and sub-tidal studies. Direct human observation is a particularly effective ground–truthing method and is most often used in inter-tidal studies where sampling sites are easily accessible. Each observer effectively takes thousands of visual ‘samples’ over a wide range of spatial scales and can make a very rapid assessment of the shore in relation to the remote survey map (e.g. an aerial photo) and can direct sampling effort (ad hoc) to those areas where it is most needed. A single ‘walk-over’ survey may be all that is required to effectively ground-truth the remote survey data. In shallow sub-tidal areas, divers can make similarly effective direct observations, though the areas they can cover are limited by mobility and visibility under water.
Examples of a variety of ground-truth techniques: 
 (photo credits: 1 & 2 JNCC, 3 & 4 Cefas)
              Walk Over Surveys
Walk over surveys of the shore
              Diver Surveys
Diver surveys in shallow water
               Survey Sled
Video Surveys
                   Survey Grab
Grab Sampling
For the deeper sub-tidal areas (usually sites away from the shore) ground truthing is often wholly reliant on using a variety of remote sampling devices such as grabs, trawls and video/stills cameras, the selection of which is discussed in Selecting a suite of tools. As sampling effort can no longer be directed ad hoc by our own observations we must rely on some a priori rules to determine sampling frequency. These are discussed further in the section on Ground-truth Survey Design.

Number of samples

Ground-truth sampling is a critical step in classifying habitats and needs to be effective but not over-burdening. The prospect of being over-burdened with samples relates to the number of samples obtained and the manner in which they are analysed. Given that acquiring samples is often the most costly part of a ground-truth survey, it would seem appropriate to ensure that each sample is thoroughly processed in order to obtain maximum long-term benefit. Detailed information can always be aggregated to suit the purpose of broadscale mapping, but the converse is not true; fine detail cannot be derived from ‘superficial’ data. In the long term it is more cost efficient to ensure that the data is collected once and used many times, rather than having to repeat sampling to satisfy different data needs.
The strategic selection of a suite of sampling techniques can reduce the overall sampling effort needed for ground-truth surveys, as many techniques provide both physical and biological information. For example, grab samples can provide information on both the nature of sediments and the composition of the biota they support. It is important to consider both the complementarity of sampling sites and sampling techniques when designing the ground-truth survey. Further details are given in the sections Selecting a suite of tools and Ground-truth Survey Design.

Peripheral information

Gound-truth surveys may also provide an opportunity to collect environmental data or information (observations) that are used as categorical descriptors of habitats, such as salinity, tidal stream and wave exposure (see table). Precise measurement of variables (e.g. tidal stream) is not usually required and the quoted bounds of the categories (e.g. “strong tides” as 1.5 to 3.0 m per second) should be regarded as a guide rather than being definitive.
                    Catagorical descriptors
Links to other chapters:
Links to other sections:

All material variously copyrighted by MESH project partners 2004-2010

| List Access Keys | Mapping European Seabed Habitats (MESH) - Home | Overview | News | Work areas in the MESH Project | Partners | Contact Us | Search | Site Map | Partner Extranet |